Producing plankton classifiers that are robust to dataset shift

Limnology and Oceanography Methods, Journal Year: 2024, Volume and Issue: unknown

Published: Nov. 27, 2024

Abstract Modern plankton high‐throughput monitoring relies on deep learning classifiers for species recognition in water ecosystems. Despite satisfactory nominal performances, a significant challenge arises from dataset shift, which causes performances to drop during deployment. In our study, we integrate the ZooLake dataset, consists of dark‐field images lake (Kyathanahally et al. 2021a), with manually annotated 10 independent days deployment, serving as test cells benchmark out‐of‐dataset (OOD) performances. Our analysis reveals instances where classifiers, initially performing well in‐dataset conditions, encounter notable failures practical scenarios. For example, MobileNet 92% accuracy shows 77% OOD accuracy. We systematically investigate conditions leading performance drops and propose preemptive assessment method identify potential pitfalls when classifying new data, pinpoint features that adversely impact classification. present three‐step pipeline: (i) identifying degradation compared performance, (ii) conducting diagnostic causes, (iii) providing solutions. find ensembles BEiT vision transformers, targeted augmentations addressing robustness, geometric ensembling, rotation‐based test‐time augmentation, constitute most robust model, call BEsT . It achieves an 83% accuracy, errors concentrated container classes. Moreover, it exhibits lower sensitivity reproduces abundances. proposed pipeline is applicable generic contingent availability suitable cells. By critical shortcomings offering procedures fortify models against study contributes development more reliable classification technologies.

Language: Английский

DeepLOKI- a deep learning based approach to identify zooplankton taxa on high-resolution images from the optical plankton recorder LOKI

Frontiers in Marine Science, Journal Year: 2023, Volume and Issue: 10

Published: Nov. 30, 2023

Zooplankton play a crucial role in the ocean’s ecology, as they form foundational component food chain by consuming phytoplankton or other zooplankton, supporting various marine species and influencing nutrient cycling. The vertical distribution of zooplankton ocean is patchy, its relation to hydrographical conditions cannot be fully deciphered using traditional net casts due large depth intervals sampled. Lightframe On-sight Keyspecies Investigation (LOKI) concentrates with that leads flow-through chamber camera taking images. These high-resolution images allow for determination taxa, often even genus level, and, case copepods, developmental stages. Each cruise produces substantial volume images, ideally requiring onboard analysis, which presently consumes significant amount time necessitates internet connectivity access EcoTaxa Web service. To enhance analyses, we developed an AI-based software framework named DeepLOKI, utilizing Deep Transfer Learning Convolution Neural Network Backbone. Our DeepLOKI can applied directly on board. We trained validated model pre-labeled from four cruises, while fifth were used testing. best-performing model, self-supervised pre-trained ResNet18 Backbone, achieved notable average classification accuracy 83.9%, surpassing regularly frequently method (default) this field factor two. In summary, tool pre-sorting black white high accuracy, will simplify quicken final annotation process. addition, provide user-friendly graphical interface efficient concise processes leading up stage. Moreover, performing latent space analysis Backbone could prove advantageous identifying anomalies such deviations image parameter settings. This, turn, enhances quality control data. methodology remains agnostic specific imaging end system used, Loki, UVP, ZooScan, long there sufficient appropriately labeled data available enable effective task performance our algorithms.

Language: Английский